Performance Fundamentals in Computer Architecture

Performance is a broad term in the computer world, dependent on the specific context.
- Gaming, networking, and processor-specific tasks each have their own performance requirements.
- The perceived performance of a computer by a user is related to how quickly it responds to commands.
Traditionally, a computer's performance was often limited by its slowest component (e.g., hard disk).
Modern performance evaluation involves benchmark tests for individual components (GPU, CPU, motherboard) and overall system performance.
- Benchmark software helps users compare hardware before purchasing.
This lecture focuses specifically on CPU time, which is the time the CPU spends executing instructions.

Response time refers to the total time it takes for a system to respond to a request.
- It includes queuing delays, transmission through layers, execution time, etc.
- This lecture's focus is limited to the CPU's execution time of machine code.
Latency is related to delay but can vary based on different parameters, such as processing load.
Throughput is another performance metric, representing the amount of work done in a unit of time.
- Increasing throughput often requires adding more resources (e.g., computers).
- Improving individual computer performance can also increase throughput.
- Throughput is not the primary focus of this lecture.
CPU time can be divided into user CPU time (for application programs) and system CPU time (for the operating system).
- This lecture will focus on user CPU time.
Modern benchmark programs can provide numerous performance metrics at once.
Improving program performance, in the context of CPU time, means reducing the execution time.

If computer A completes a task in 20 seconds and computer B takes 25 seconds, computer A is more performant.
- $\frac{20}{25} = 0.8$ , inverting gives 1.25, meaning A is 1.25 times faster than B.
When analyzing CPU performance, several key factors are considered:
- How quickly a command is executed, which involves fetching, decoding, and executing.
- The role of the clock signal in synchronizing operations within the CPU.
- The CPU's clock frequency, often measured in gigahertz (GHz).
- $1 \text{ GHz} = 1 \times 10^9 \text{ cycles per second}$
A CPU with a clock frequency of 1 GHz has a clock cycle time (period) of 1 nanosecond.

Not every instruction executes in a single clock cycle.
The number of cycles per instruction (CPI) varies depending on the instruction type and processor architecture.
CPI represents the average number of clock cycles required to execute an instruction.
To calculate the execution time, you need to know:
- The number of instructions.
- The average CPI.
- The clock cycle time (period).
$\text{Execution Time} = \text{Number of Instructions} \times \text{CPI} \times \text{Clock Cycle Time}$

The architecture of the processor and the instruction set architecture (ISA) impact performance.
Different ISAs have different instruction types, which may take varying numbers of cycles to execute.
Compilers can optimize code, potentially reducing the number of instructions needed.
Three main parameters affect performance:
- Number of instructions.
- CPI.
- Clock frequency.

To compare the performance of two computers (A and B), you can measure their CPU times.
$\text{Performance Ratio} = \frac{\text{CPU Time 1}}{\text{CPU Time 2}}$
A smaller CPU time indicates better performance.
Key factors influencing CPU time include CPI, instruction count, and clock frequency.
Modern processors can dynamically adjust their clock frequencies, especially in mobile devices.

Increasing clock frequency doesn't always guarantee increased performance.
A processor's performance also depends on how well it serves multiple programs and users.
Other factors, such as the materials used in the processor and its internal features, can influence performance.
When selecting hardware, consider benchmark results and performance tests, particularly in graphics-intensive applications.
Pay attention to both CPU and GPU performance, as well as the overall system performance.
Modern benchmark programs offer comprehensive performance reports.

The lecture references a video with detailed examples of performance calculations.
It mentions older SPEC benchmark results but notes that they are outdated.
It advises students not to focus too much on the older performance criteria.
It recommends researching modern CPU benchmark software when purchasing or upgrading hardware.
The lecture highlights the adaptive nature of modern processors, which can adjust their frequencies based on power and performance needs.